Abnormality detection device of fuel vapor escape prevention system
At the time of stopping operation of the vehicle, the pressures inside of the fuel tank (5) and inside of the canister (6) detected at every constant time are stored in the storage device. A learned neural network using the pressures inside the fuel tank (5) and inside the canister (6) for each fixed time stored in the storage device and the atmospheric pressure as input parameters of the neural network and using a case where perforation occurs in the system causing fuel vapor to leak as a truth label is stored. At the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak is detected from these input parameters by using the learned neural network.
Latest Toyota Patents:
The present invention relates to an abnormality detection device of a fuel vapor escape prevention system.
BACKGROUNDIn an internal combustion engine, to prevent fuel vapor from escaping to the outside atmosphere, in the past use has been made of a fuel vapor escape prevention system comprising a canister formed with a fuel vapor chamber and atmospheric pressure chamber at the two sides of an activated carbon layer, making the fuel vapor chamber on the one hand communicate with an inside space above a fuel level of a fuel tank and on the other hand connecting it through a purge control valve to an intake passage of the engine. In such a fuel vapor escape prevention system, for example, if the walls of a fuel vapor flow pipe connecting the fuel vapor chamber and purge control valve of the canister are perforated by a hole, fuel vapor will end up escaping through the hole to the outside atmosphere.
Therefore, known in the art is a diagnosis device designed to detect a pressure at the inside space above the fuel level at the inside of the fuel tank (simply referred to as the “pressure inside the fuel tank”) and diagnose if an abnormality has occurred in the fuel vapor escape prevention system from a change of the pressure inside the fuel tank, for example, if the walls of the fuel vapor flow pipe are perforated by a hole (for example, see Japanese Unexamined Patent Publication No. 2004-44396).
In this diagnosis device, by making the purge control valve open in the state where the atmospheric pressure chamber of the canister is cut off from the atmosphere when the vehicle is being driven steadily and stably, the pressure inside the fuel tank is lowered below the atmospheric pressure, then by making the purge control valve close, the inside of the fuel tank is made to be in a scaled state. At this time, for example, if the walls of the fuel vapor flow pipe are perforated by a hole, the pressure inside the fuel tank will rise a little at a time. Therefore, in this diagnosis device, when the pressure inside the fuel tank increases more than a fixed value after the inside of the fuel tank is made in a sealed state, for example, it is judged that the walls of the fuel vapor flow pipe are perforated by a hole.
SUMMARYIn this case, if the diameter of the hole formed in the walls of the fuel vapor flow pipe is small, the amount of rise of the pressure inside the fuel tank will become small. On the other hand, the pressure inside the fuel tank also fluctuates due to other factors, for example, the temperature of the fuel inside the fuel tank. Therefore, if it is judged that the walls of the fuel vapor flow pipe are perforated by a hole when the pressure inside the fuel tank increases more than a fixed value, there is the risk of mistaken judgment.
The present invention provides an abnormality detection device of a fuel vapor escape prevention system using a neural network to, for example, detect perforation of the walls of a fuel vapor flow pipe and able to accurately detect such perforation of the walls of the fuel vapor flow pipe even if, at this time, the diameter of the hole is small.
That is, according to the present invention, there is provided an abnormality detection device of a fuel vapor escape prevention system comprising:
-
- a canister formed with a fuel vapor chamber and atmospheric pressure chamber at the two sides of an activated carbon layer, the fuel vapor chamber being on the one hand communicated with an inside space above a fuel level of a fuel tank and on the other hand communicated through a purge control valve with an intake passage of an engine,
- a flow path switching valve able to selectively connect the atmospheric pressure chamber to the atmosphere and a suction pump, and
- a pressure sensor detecting pressure at an inside of the fuel tank and inside of the canister, wherein
- at the time of stopping operation of the vehicle, processing for detection of an abnormality is performed to generate a valve closing instruction making the purge control valve close, a switching instruction switching a switched position of the flow path switching valve to a switched position at which the atmospheric pressure chamber is connected to the suction pump, and a pump operation instruction making the suction pump operate to make the inside of the fuel tank and inside of the canister a negative pressure,
- at the time the processing for detection of an abnormality is performed, a pressures at the inside of the fuel tank and inside of the canister detected by the pressure sensor at every fixed time are stored in a storage device,
- a learned neural network learned in weights using the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the system causing leakage of fuel vapor as a truth label is stored, and
- at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak is detected from the above mentioned input parameters by using the learned neural network.
Furthermore, according to the present invention, there is provided an abnormality detection device of a fuel vapor escape prevention system comprising:
-
- a canister formed with a fuel vapor chamber and atmospheric pressure chamber at the two sides of an activated carbon layer, the fuel vapor chamber being on the one hand communicated with an inside space above a fuel level of a fuel tank and on the other hand communicated through a purge control valve with an intake passage of an engine,
- a flow path switching valve able to selectively connect the atmospheric pressure chamber to the atmosphere and a suction pump, a passage from the flow path switching valve toward the atmospheric pressure chamber and a suction passage from the flow path switching valve toward the suction pump being connected by a reference pressure detection passage having a restricted opening, and
- a pressure sensor arranged in the suction passage from the flow path switching valve toward the suction pump,
- at the time of stopping operation of the vehicle, processing for detection of an abnormality is performed to generate a valve closing instruction making the purge control valve close, a pump operation instruction making the suction pump operate to make an inside of the fuel tank and inside of the canister a negative pressure while maintaining a switched position of the flow path switching valve at a switched position where the atmospheric pressure chamber is connected to the atmosphere when a predetermined time elapses after stopping operation of the vehicle, a switching instruction switching the switched position of the flow path switching valve to a switched position at which the atmospheric pressure chamber is connected to the suction pump after generation of the pump operation instruction, and a valve opening instruction making the purge control valve open after the generation of the switching instruction,
- at the time the processing for detection of an abnormality is performed, a pressures at the inside of the fuel tank and inside of the canister detected by the pressure sensor at every fixed time are stored in a storage device,
- a learned neural network learned in weights using the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the system causing leakage of fuel vapor as a truth label is stored, and,
- at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak is detected from the above mentioned input parameters by using the learned neural network.
By using the pressures at the inside of the fuel tank and inside of the canister detected by the pressure sensor at every fixed time and at least the atmospheric pressure as input parameters of the neural network for learning the weights of the neural network, for example, it is possible to accurately detect perforation of the walls of the fuel vapor flow pipe even if a small diameter hole is formed in the walls of the fuel vapor flow pipe.
Overall Configuration of Internal Combustion Engine
As shown in
In
In a hybrid engine provided with an electric motor as a drive source, if the driving start/stop switch 31 is set to ON, operation of the vehicle by the engine or electric motor is started, while if the driving start/stop switch 31 is set to OFF, operation of the vehicle by the engine or electric motor is stopped. On the other hand, in an engine not provided with an electric motor as a drive source, if the driving start/stop switch 31 is set to ON, the engine is started and operation of the vehicle is started, while if the driving start/stop switch 31 is set to OFF, the engine is stopped and operation of the vehicle is stopped.
Further, as shown in
Inside the flow path switching valve 42, a first passage 48 able to connect the atmospheric pressure chamber connection path 43 and the atmosphere communication path 45 as shown in
On the other hand, when detecting an abnormality of the fuel vapor escape prevention system, the flow path switching valve 42 is switched to a position at which the atmospheric pressure chamber connection path 43 is connected to the suction passage 46 through the second passage 49 as shown in
Next, referring to
Gravity acts on the weight 64 in the vertical direction, so force downward in the vertical direction acts at the center of the cross-shaped sheet 63. At this time, if the frame 61 is positioned in a horizontal plane, the amounts of strain of the arms 62 are the same. Therefore, when the vehicle is stopped on a horizontal surface, the amounts of strain of the arms 62 become the same. As opposed to this, if the vehicle is stopped inclined with respect to the horizontal plane, the amounts of strain of the arms 62 become different values. The direction of inclination and amount of inclination of the vehicle with respect to the horizontal plane can be learned from the differences in amounts of strain of the arms 62. On the other hand, the height of the fuel level inside of the fuel tank 5 when the vehicle is positioned on a flat surface can be learned from the detected value of the fuel level gauge 7.
Therefore, no matter what shape the fuel tank 5, the remaining amount of fuel in the fuel tank 5 can be learned from the detected value of the fuel level gauge 7 and the direction of inclination and amount of inclination of the vehicle with respect to the horizontal plane detected by the gravity sensor 60. Therefore, in this example, as shown in
Next, the basic thinking of the present invention will be explained. As will be understood from
Now then, in
On the other hand, a dash dot line of
However, if the walls of the fuel vapor circulation pipe 12 or 13 are perforated by a hole, that is, in the case shown by the broken line, compared with the cases shown by the solid line and the dash and dot line, the degree of drop of pressure will become smaller in the middle of the drop in pressure, and the overall shape of the curve of the drop differs between the case shown by the broken line and the cases shown by the solid line and the dash and dot line. Therefore, if finding the overall shape of the curve of the pressure drop, it becomes possible to accurately judge if the walls of the fuel vapor flow pipe 12 or 13 are perforated by a hole from the differences in the overall shape of the curve of the pressure drop. Therefore, in the present invention, the system internal pressure is detected at every fixed time, and a neural network is used to judge if an abnormality occurs in the fuel vapor escape prevention system based on the system internal pressure detected at every fixed time.
Summary of Neural Network
As explained above, in the embodiment according to the present invention, a neural network is used to judge if an abnormality occurs in the fuel vapor escape prevention system. Therefore, first, a neural network will be briefly explained.
At the nodes of the input layer, the inputs are output as they are. On the other hand, the output values x1 and x2 of the nodes of the input layer are input at the nodes of the hidden layer (L=2), while the respectively corresponding weights “w” and biases “b” are used to calculate sum input values “u” at the nodes of the hidden layer (L=2). For example, a sum input value uk calculated at a node shown by z(2)k (k=1, 2, 3) of the hidden layer (L=2) in
Next, this sum input value uk is converted by an activation function “f” and is output from a node shown by z(2)k of the hidden layer (L=2) as an output value z(2)k (=f(uk)). On the other hand, the nodes of the hidden layer (L=3) receive as input the output values z(2)1, z(2)2, and z(2)3 of the nodes of the hidden layer (L=2). At the nodes of the hidden layer (L=3), the respectively corresponding weights “w” and biases “b” are used to calculate the sum input values “u” (Σz·w+b). The sum input values “u” are similarly converted by an activation function and output from the nodes of the hidden layer (L=3) as the output values z(3)1, z(3)2, and z(3)3. As this activation function, for example, a Sigmoid function σ is used.
On the other hand, at the nodes of the output layer (L=4), the output values z(3)1, z(3)2, and z(3)3 of the nodes of the hidden layer (L=3) are input. At the nodes of the output layer, the respectively corresponding weights “w” and biases “b” are used to calculate the sum input values “u” (Σz·w+b) or just the respectively corresponding weights “w” are used to calculate the sum input values “u” (Σz·w). For example, at the regression problem, at the nodes of the output layer, an identity function is used, therefore, from the nodes of the output layer, the sum input values “u” calculated at the nodes of the output layer are output as they are as the output values “y”.
Learning in Neural Network
Now then, if designating the teacher data showing the truth values of the output values “y” of the neural network, that is, the truth data, as yt, the weights “w” and biases “b” in the neural network are learned using the error backpropagation algorithm so that the difference between the output values “y” and the teacher data, that is, the truth data yt, becomes smaller. This error backpropagation algorithm is known. Therefore, the error backpropagation algorithm will be explained simply below in its outlines. Note that, a bias “b” is one kind of weight “w”, so below, a bias “b” will be also be included in what is referred to as a weight “w”. Now then, in the neural network such as shown in
[Equation 2]
∂E/∂w(L)=(∂E/∂u(L))(∂u(L)/∂w(L)) (1)
where, z(L−1)·∂w(L)=∂u(L) so if (∂E/∂u(L)=δ(L), the above equation (1) can be shown by the following equation:
[Equation 3]
∂E/∂w(L)=δ(L)·z(L−1) (2)
where, if u(L) fluctuates, fluctuation of the error function E is caused through the change in the sum input value u(L+1) of the following layer, so δ(L) can be expressed by the following equation:
where, if expressing z(L)=f(u(L)), the input value uk(L+1) appearing at the right side of the above equation (3) can be expressed by the following formula:
where, the first term (∂E/∂u(L+1)) at the right side of the above equation (3) is δ(L+1), and the second term (∂uk(L+1)/∂u(L)) at the right side of the above equation (3) can be expressed by the following equation:
[Equation 6]
∂(wk(L+1)·z(L)/∂u(L)=wk(L+1)·∂f(u(L))/∂u(L)=wk(L+1)·f(u(L)) (5)
Therefore, δ(L) is shown by the following formula.
That is, if δ(L+1) is found, δ(L) can be found.
Now then, if there is a single node of the output layer (L=4), teacher data, that is, truth data yt, is found for a certain input value, and the output values from the output layer corresponding to this input value are “y”, if the square error is used as the error function, the square error E is found by E=1/2(y−yt)2. In this case, at the node of the output layer (L=4), the output values “y” become f(u(L)), therefore, in this case, the value of δ(L) at the node of the output layer (L=4) becomes like in the following equation:
[Equation 8]
δ(L)=∂E/∂u(L)=(∂E/∂y)(∂y/∂u(L))=(y−yt)·f(u(L)) (7)
In this case, in the regression problem, as explained above, f(u(L)) is an identity function and f(u(L1))=1. Therefore, this leads to δ(L)=y−yt and δ(L) is found.
If δ(L) is found, the above equation (6) is used to find the δ(L−1) of the previous layer. The δ's of the previous layer are successively found in this way. Using these values of δ's, from the above equation (2), the differential of the error function E, that is, the slope ∂E/∂w(L), is found for the weights “w”. If the slope ∂E/∂w(L) is found, this slope ∂E/∂w(L) is used to update the weights “w” so that the value of the error function E decreases. That is, the weights “w” arc learned.
On the other hand, in the classification problem, at the time of learning, the output values y1, y2 . . . from the output layer (L=4) are input to a softmax layer. If defining the output values from the softmax layer as y1′, y2′ . . . and the corresponding truth labels as yt1, yt2 . . . as the error function E, the following cross entropy error E is used.
In this case as well, the values of δ(L) at the nodes of the output layer (L=4) become δ(L)=yk−ytk (k=1, 2 . . . n). From the values of these δ(L), the δ(L−1) of the previous layers are found using the above equation (6).
Embodiments of Present InventionFirst, referring to
In
Next, if reaching the time t1, the switching instruction switching the flow path switching valve 42 to the test position and the instruction for operating the suction pump 40 are issued. On the other hand, the valve closing instruction continues to be issued to the purge control valve 14. At this time, the inside of the fuel tank 5, the inside of the canister 6, the inside of the fuel vapor flow pipe 12, and the inside of the fuel vapor flow pipe 13 between the canister 6 and the purge control valve 14 form a sealed space isolated from the atmosphere. If the suction pump 40 is made to operate in such a state, the air inside this sealed space is gradually sucked in by the suction pump 40. As a result, the pressure at the inside of the fuel tank 5 and the inside of the canister 6, that is, the system internal pressure detected by the pressure sensor 47, gradually falls. Next, around when reaching the time t2, no pressure drop occurs any longer and the system internal pressure hovers at the dropped state.
If reaching the time t2, the valve opening instruction making the purge control valve 14 open is issued. On the other hand, at this time, the flow path switching valve 42 is maintained at the test position and the suction pump 40 continues to be operated. Therefore, at this time, the suction action by the suction pump 40 is continued, but the purge control valve 14 is made to open, so the system internal pressure rapidly rises and becomes atmospheric pressure. Next, if reaching the time t3, the purge control valve 14, flow path switching valve 42, and suction pump 40 are returned to the states at the time of stopping operation of the vehicle. That is, if reaching the time t3, the valve closing instruction making the purge control valve 14 close, the switching instruction switching the flow path switching valve 42 to the normal position, and the instruction for stopping the suction pump 40 are issued.
On the other hand, if an abnormality occurs in the fuel vapor escape prevention system, the pattern of change in the system internal pressure detected by the pressure sensor 47 becomes a pattern of change different from the pattern of change in the system internal pressure at the normal time shown in
The solid line in
On the other hand, the solid line of
In this way, if an abnormality occurs, the pattern of change in the system internal pressure becomes a pattern of change different from the pattern of change in the system internal pressure at the normal time. Therefore, in the embodiment according to the present invention, to learn the differences in the pattern of change in the system internal pressure, the system internal pressure is detected at every fixed time. Next, this will be explained while referring to
Referring to
Next, the neural network used for preparing the abnormality judgment estimation model will be explained while referring to
On the other hand,
Next, the input values xx1, xx2 . . . xxk−1, and xxk in
As shown in
On the other hand, when the system internal pressure becomes a negative pressure due to the suction action by the suction pump 40, if the fuel in the fuel tank 5 evaporates, the system internal pressure rises. In this case, the greater the amount of evaporation of the fuel per unit time, the greater the amount of change in the system internal pressure. On the other hand, the amount of evaporation of the fuel per unit time is proportional to the remaining amount of fuel in the fuel tank 5. Therefore, the greater the remaining amount of fuel in the fuel tank 5, the greater the influence given to the system internal pressure. Therefore, as shown in
Further, if the temperature of the fuel in the fuel tank 5 rises, the amount of evaporation of the fuel per unit time increases, so the temperature of the fuel in the fuel tank 5 also has an influence on the system internal pressure. However, this influence on the system internal pressure is smaller than the remaining amount of fuel in the fuel tank 5, so as shown in
Now then, as the input values x1, x2 . . . xn−1, and xn and the input values xx1, xx2 . . . xxk−1, and xxk of the neural network 70 shown in
On the other hand, in
In this case, for example, when a perforation abnormality occurs in which the walls of the vapor circulation pipe 12 or 13 are perforated by a small hole, only the truth label yt1 is made 1, while the remaining truth labels yt2, yt3, and yt4 are all made zero. Similarly, when a valve opening abnormality occurs in which the purge control valve 14 continues opened, only the truth label yt2 is made 1, while the remaining truth labels yt1, yt3, and yt4 are all made zero. When a valve closing abnormality occurs in which the purge control valve 14 continues closed, only the truth label yt3 is made 1, while the remaining truth labels yt1, yt2, and yt4 are all made zero, and when normal, only the truth label yt4 is made 1 and the remaining truth labels yt1, yt2, and yt3 are all made zero.
On the other hand, as shown in
Next, the method of preparation of the training data set shown in
While this processing for detection of an abnormality is being performed, the data required for preparing the training data set is acquired.
Referring to
On the other hand, when at step 101 it is judged that the time “t” is not the time t1 shown in
On the other hand, when at step 106 it is judged that the time “t” is not the time t2 shown in
In this way, the system internal pressure xn for each fixed time Δt when the combination of the atmospheric pressure, the remaining amount of fuel in the fuel tank 5, the temperature of the fuel in the fuel tank 5, and the characteristic value of flow rate of the suction pump 40 is changed in each state of a state where a perforation abnormality occurs in which the walls of the fuel vapor flow pipe 12 or 13 are perforated by a small hole, a state where a valve opening abnormality occurs in which the purge control valve 14 continues opened, a state where a valve closing abnormality occurs in which the purge control valve 14 continues closed, and a normal state are stored in the test control device 81. That is, the No. 1 to No. “m” input values x1m, x2m . . . xnm−1, and xnm, the input values xx1m, xx2m . . . xxkm−1, and xxkm, and the truth labels ytsm (m=1, 2, 3 . . . m) of the training data set shown in
If a training data set such as shown in
Next, at step 203, the weights of the neural network 70 are learned. At this step 203, first, the No. 1 input values x1, x2 . . . xn−1, and xn and input values xx1, xx2 . . . xxk−1, and xxk of
If the weights of the neural network 70 finish being learned based on the No. 1 data of
At step 204, it is judged if the cross entropy error E becomes a preset error setting or less. When it is judged that the cross entropy error E does not become the preset error setting or less, the routine returns to step 203 where, again, learning of the weights of the neural network 70 is performed based on the training data set shown in
In the embodiment according to the present invention, the thus prepared abnormality judgment estimation model of the fuel vapor escape prevention system is used to diagnose a fault in the fuel vapor escape prevention system of a commercially available vehicle. To this end, the abnormality judgment estimation model of the fuel vapor escape prevention system is stored in the electronic control unit 20 of the commercially available vehicle.
If referring to
Next, referring to
If this processing for detection of an abnormality is performed, the system internal pressure xn is acquired for at every fixed time Δt and the system internal pressure xn acquired at every fixed time Δt is stored in the memory 22 of the electronic control unit 20. Next, at step 403, it is judged if the processing for detection of an abnormality has ended. When the processing for detection of an abnormality has not ended, the processing cycle is ended. As opposed to this, when it is judged that the processing for detection of an abnormality has ended, the routine proceeds to step 404 where the detection permission flag is set. If the detection permission flag is set, at the next processing cycle, the routine proceeds from step 401 to step 405.
At step 405, the system internal pressures x1, x2 . . . xn−1, and xn for each fixed time Δt stored in the memory 22 of the electronic control unit 20 are read in. Next, at step 406, the input values xx1, xx2 . . . xxk−1, and xxk stored in the memory 22 of the electronic control unit 20 are read in. Next, at step 407, the system internal pressures xnx1, x2 . . . xn−1, and xn for each fixed time Δt and the input values xx1, xx2 . . . xxk−1, and xxk are input to the nodes of the input layer (L=1) of the neural network 71 shown in
Next, at step 409, the largest output value y1′ is selected from the acquired output values y1′, y2′, y3′, and y4′. At this time, it is estimated that the abnormal state shown in
In this way, in the abnormality detection device of a fuel vapor escape prevention system according to the present invention, the fuel vapor escape prevention system is provided with the canister 6 at which the fuel vapor chamber 10 and atmospheric pressure chamber 11 are formed at the two sides of the activated carbon layer 9. The fuel vapor chamber 10 is on the one hand communicated with the inside space above the fuel level of the fuel tank 5 and is on the other hand communicated with the inside of the intake passage of the engine through the purge control valve 14. Furthermore, the fuel vapor escape prevention system is provided with the flow path switching valve 42 able to selectively connect the atmospheric pressure chamber 11 to the atmosphere and the suction pump 40 and the pressure sensor 47 detecting the pressure at the inside of the fuel tank 5 and the inside of the canister 6. At the time of stopping operation of the vehicle, processing for detection of an abnormality generating a valve closing instruction making the purge control valve 14 close, a switching instruction switching the switched position of the flow path switching valve 42 to the switched position where the atmospheric pressure chamber 11 is connected to the suction pump 40, and a pump operation instruction making the suction pump 40 operate so as to make the inside of the fuel tank and the inside of the canister 6 a negative pressure is performed. When this processing for detection of an abnormality is performed, the pressure at the inside of the fuel tank 5 and the inside of the canister 6 detected at every fixed time by the pressure sensor 40 is stored in the storage device. The learned neural network learned in weights using the pressure at the inside of the fuel tank 5 and inside of the canister 6 at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the system causing leakage of fuel vapor as a truth label is stored, and, at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak is detected from the input parameters by using the learned neural network.
In this case, in the embodiment according to the present invention, the above-mentioned processing for detection of an abnormality includes processing for generating a valve opening instruction making the purge control valve 14 open after generating the valve closing instruction of the purge control valve 14, a learned neural network learned in weights using the pressure at the inside of the fuel tank 5 and inside of the canister 6 at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the above-mentioned system causing leakage of fuel vapor, a case where a valve opening abnormality occurs in which the purge control valve 14 continues opened, and a case where a valve closing abnormality occurs in which the purge control valve 14 continues closed as truth labels is stored, and, at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak, a valve opening abnormality of the purge control valve 14, and a valve closing abnormality of the purge control valve 14 are detected from the input parameters by using the learned neural network.
Further, in this embodiment according to the present invention, the above-mentioned input parameters are comprised of the pressures at the inside of the fuel tank 5 and the inside of the canister 6 at every fixed time stored in the storage device and the atmospheric pressure when processing for detection of an abnormality is performed and the remaining amount of fuel in the fuel tank 5 when processing for detection of an abnormality is performed. Alternatively, in the embodiment according to the present invention, the above-mentioned input parameters are comprised of the pressures at the inside of the fuel tank 5 and the inside of the canister 6 at every fixed time stored in the storage device and the atmospheric pressure when processing for detection of an abnormality is performed, the remaining amount of fuel in the fuel tank 5 when processing for detection of an abnormality is performed, the temperature of the fuel of the fuel tank 5, and a parameter showing the capacity of the suction pump 40.
Next, processing for detection of an abnormality performed using the suction pump module 16 shown in
In
Next, if reaching the time t2, an instruction for operating the suction pump 40 is issued. At this time, the flow path switching valve 42 has been switched to the normal position shown in
In this case, the changes in the system internal pressure detected by the pressure sensor 47 show the changes in the system internal pressure when perforated by a hole of the same diameter as the diameter of the restricted opening 51. Therefore, the changes in the system internal pressure at this time become the reference for judging whether the fuel vapor escape prevention system is perforated by a hole. Therefore, the passage 50 will be referred to as the “reference pressure detection passage”. Next, if reaching the time t2, a switching instruction switching the flow path switching valve 42 to the test position shown in
If reaching the time t4, a valve opening instruction making the purge control valve 14 open is issued. On the other hand, at this time, the flow path switching valve 42 is maintained at the test position and the suction pump 40 continues to be operated. Therefore, at this time, the suction action due to the suction pump 40 is continued, but the purge control valve 14 is made to open, so the system internal pressure rapidly rises and becomes atmospheric pressure. Next, if reaching the time t5, the purge control valve 14, flow path switching valve 42, and suction pump 40 are returned to the states at the time of stopping operation of the vehicle. That is, if reaching the time t5, a valve closing instruction making the purge control valve 14 close, a switching instruction switching the flow path switching valve 42 to the normal position, and an instruction for stopping the suction pump 40 are issued.
On the other hand, in this embodiment, as shown in
In this embodiment as well, if an abnormality occurs in the fuel vapor escape prevention system, the pattern of change in the system internal pressure detected by the pressure sensor 47 becomes a pattern of change different from the pattern of change in the system internal pressure at the normal time shown in
The solid line of
The solid line of
The solid line of
The solid line of
The solid line of
The solid line of
The solid line of
The solid line in
If an abnormality occurs in this way, the pattern of change in the system internal pressure becomes a pattern of change different from the pattern of change in the system internal pressure at the normal time. Therefore, in this embodiment as well, as shown in
Next, the neural network 72 used for preparation of the abnormality judgment estimation model will be explained while referring to
On the other hand,
The input values xx1, xx2 . . . xxk−1, and xxk in
Now then, in this embodiment as well, as the input values x1, x2 . . . xn−1, and xn and the input values xx1, xx2 . . . xxk−1, and xxk of the neural network 72 shown in
In this embodiment as well, first, the input values x1, x2 . . . xn−1, and xn, the input values xx1, xx2 . . . xxk−1, and xxk, and the training data, that is, the truth labels yt, are used to prepare the training data set shown in
On the other hand, in
In this case, for example, when a perforation abnormality occurs in which the walls of the vapor flow pipe 12 or 13 are perforated by a small hole, only the truth label yt1 is made 1, while the remaining truth labels yt2, yt3, yt4, yt5, yt6, yt, yt8, and yt9 are all made zero. Similarly, when a valve opening abnormality occurs in which the purge control valve 14 continues opened, only the truth label yt2 is made 1, while the remaining truth labels yt1, yt3, yt4, yt5, yt6, yt7, yt8, and yt9 are all made zero. When a valve closing abnormality occurs in which the purge control valve 14 continues closed, only the truth label yt3 is made 1, while the remaining truth labels yt1, yt2, yt4, yt5, yt6, yt7, yt8, and yt9 are all made zero. When an abnormality occurs in the pressure sensor 47, only the truth label yt4 is made 1, while the remaining truth labels yt1, yt2, yt3, yt5, yt6, yt7, yt8, and yt9 are all made zero. When an abnormality occurs in which the flow path switching valve 42 sticks at the normal position, only the truth label yt5 is made 1, while the remaining truth labels yt1, yt2, yt3, yt4, yt6, yt7, yt8, and yt9 are all made zero. When an abnormality occurs in which the flow path switching valve 42 sticks at the test position, only the truth label yt6 is made 1, while the remaining truth labels yt1, yt2, yt3, yt4, yt5, yt7, yt8, and yt9 are all made zero. When an abnormality occurred in which the suction pump 40 continues operating, only the truth label yt7 is made 1, while the remaining truth labels yt1, yt2, yt3, yt4, yt5, yt6, yt8, and yt9 are all made zero. When an abnormality occurs in which the suction pump 40 continues to stop operating, only the truth label yt8 is made 1, while the remaining truth labels yt1, yt2, yt3, yt4, yt5, yt6, yt7, and yt9 are all made zero. When at the normal time, only the truth label yt9 is made 1, while the remaining truth labels yt1, yt2, yt3, yt4, yt5, yt6, yt7, and yt8 are all made zero.
On the other hand, as shown in
This training data set is also prepared by a method similar to the method already explained with reference to
While this processing for detection of an abnormality is being performed, the data required for preparing the training data set is acquired.
Referring to
On the other hand, when at step 501 it is judged that the time “t” is the time t1 shown in
On the other hand, when, at step 504, it is judged that the time “t” is the time t5 shown in
In this way, the system internal pressure xn for each fixed time Δt when the combination of the atmospheric pressure, the remaining amount of fuel in the fuel tank 5, the temperature of the fuel in the fuel tank 5, and the characteristic value of flow rate of the suction pump 40 is changed in each state of a state where a perforation abnormality occurs in which the walls of the fuel vapor flow pipe 12 or 13 are perforated by a small hole, a state where a valve opening abnormality occurs in which the purge control valve 14 continues opened, a state where a valve closing abnormality occurs in which the purge control valve 14 continues closed, a state where an abnormality occurs in the pressure sensor 47, a state where an abnormality of the flow path switching valve 42 sticking at the normal position occurs, a state where an abnormality occurs in which the flow path switching valve 42 sticks at the test position, a state where an abnormality occurs in which the suction pump 40 continues operating, a state where an abnormality occurs in which the suction pump 40 continues to stop operating, and a normal state are stored in the test control device 81. That is, the No. 1 to No. “m” input values x1m, x2m . . . xnm−1, and xnm, the input values xx1m, xx2m . . . xxkm−1, and xxkm, and the truth label ytsm (m=1, 2, 3 . . . m) of the training data set shown in
If a training data set is prepared in this way, electronic data of the prepared training data set is used to learn the weights of the neural network 72 shown in
In this embodiment as well, the thus prepared abnormality judgment estimation model of the fuel vapor escape prevention system is used to diagnose a fault of the fuel vapor escape prevention system in a commercially available vehicle. To this end, this abnormality judgment estimation model of the fuel vapor escape prevention system is stored in the electronic control unit 20 of the commercially available vehicle using a routine for reading data into the electronic control unit shown in
In this embodiment as well, as the routine for detection of an abnormality of the fuel vapor escape prevention system performed at a commercially available vehicle, the routine shown in
That is, at step 407, system internal pressures xnx1, x2 . . . xn−1, and xn for each fixed time Δt and input values xx1, xx2 . . . xxk−1, and xxk are input to the nodes of the input layer (L=1) of the neural network 73 shown in
Next, at step 409, the largest output value yi′ is selected from among the acquired output values y1′, y2′, y3′, y4′, y5′, y6′, y7′, y8′, and y9′. At this time, it is estimated that the abnormal state shown in
In this way, in an abnormality detection device of a fuel vapor escape prevention system according to another embodiment of the present invention, the fuel vapor escape prevention system is provided with the canister 6 at which the fuel vapor chamber 10 and the atmospheric pressure chamber 11 are formed at the two sides of the activated carbon layer 9. The fuel vapor chamber 10 is on the one hand communicated with the inside space above the fuel level of the fuel tank 5 and is on the other hand communicated through the purge control valve 14 with the intake passage of the engine. Furthermore, the fuel vapor escape prevention system is provided with the flow path switching valve 42 able to selectively connect the atmospheric pressure chamber 11 with the atmosphere and suction pump 40. The passage 43 from the flow path switching valve 42 toward the atmospheric pressure chamber 11 and the suction passage 46 from the flow path switching valve 42 toward the suction pump 40 are connected by the reference pressure detection passage 50 having the restricted opening 51. Inside the suction passage 46 from the flow path switching valve 42 toward the suction pump 40, the pressure sensor 47 is arranged. At the time of stopping operation of the vehicle, processing for detection of an abnormality is performed generating a valve closing instruction causing the purge control valve 14 to close, a pump operation instruction making the suction pump 40 operate to make the inside of the fuel tank 5 and inside of the canister 6 a negative pressure while maintaining the switched position of the flow path switching valve 42 at a switched position where the atmospheric pressure chamber 11 is connected to the atmosphere when a predetermined time elapses after stopping operation of the vehicle, a switching instruction switching the switched position of the flow path switching valve 42 after generation of the pump operation instruction to a switched position at which the atmospheric pressure chamber 11 is connected to the suction pump 40, and a valve opening instruction making the purge control valve 14 open after the generation of the switching instruction. At the time the processing for detection of an abnormality is performed, the pressure at the inside of the fuel tank 5 and inside of the canister 6 detected by the pressure sensor 47 at every fixed time are stored in the storage device, a learned neural network learned in weights using the pressures at the inside of the fuel tank 5 and inside of the canister 6 at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the system causing leakage of fuel vapor as a truth label is stored, and, at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak is detected from the input parameters by using the learned neural network.
In this case, in this embodiment according to the present invention, a learned neural network learned in weights using the pressures at the inside of the fuel tank 5 and inside of the canister 6 at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the fuel vapor escape prevention system causing leakage of fuel vapor, a case where a valve opening abnormality occurs in which the purge control valve 14 continues opened, a case where a valve closing abnormality occurs in which the purge control valve 14 continues closed, a case where an abnormality occurs in the pressure sensor 47, a case where a switching abnormality occurs in which the switched position of the flow path switching valve 42 is maintained at a switched position connecting the atmospheric pressure chamber 11 to the atmosphere, a case where a switching abnormality occurs in which the switched position of the flow path switching valve 42 is maintained at a switched position connecting the atmospheric pressure chamber 11 to the suction pump 40, a case where an abnormality occurs in which the suction pump 40 continues operating, and a case where an abnormality occurs in which the suction pump 40 continues stopped as truth labels is stored, and, at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak, a valve opening abnormality of the purge control valve, a valve closing abnormality of the purge control valve, an abnormality of the pressure sensor, a switching abnormality of the flow path switching valve, and an abnormality of the suction pump are detected from the input parameters by using the learned neural network.
Furthermore, in this case, in this embodiment according to the present invention, the above-mentioned input parameters are comprised of the pressures at the inside of the fuel tank 5 and inside of the canister 6 at every fixed time stored in the storage device, the atmospheric pressure when processing for detection of an abnormality is performed, and the remaining amount of fuel in the fuel tank 5 when processing for detection of an abnormality is performed. Further, in this case, in this embodiment according to the present invention, the above-mentioned input parameters are comprised of the pressures at the inside of the fuel tank 5 and inside of the canister 6 at every fixed time stored in the storage device, the atmospheric pressure when processing for detection of an abnormality is performed, the remaining amount of fuel in the fuel tank 5 when processing for detection of an abnormality is performed, the temperature of the fuel in the fuel tank 5, and a parameter showing the capacity of the suction pump 40.
Claims
1. An abnormality detection device of a fuel vapor escape prevention system comprising:
- a canister formed with a fuel vapor chamber and atmospheric pressure chamber at the two sides of an activated carbon layer, the fuel vapor chamber being on the one hand communicated with an inside space above a fuel level of a fuel tank and on the other hand communicated through a purge control valve with an intake passage of an engine,
- a flow path switching valve able to selectively connect the atmospheric pressure chamber to the atmosphere and a suction pump, and
- a pressure sensor detecting pressure at an inside of the fuel tank and inside of the canister, wherein
- at the time of stopping operation of the vehicle, processing for detection of an abnormality is performed to generate a valve closing instruction making the purge control valve close, a switching instruction switching a switched position of the flow path switching valve to a switched position at which the atmospheric pressure chamber is connected to the suction pump, and a pump operation instruction making the suction pump operate to make the inside of the fuel tank and inside of the canister a negative pressure,
- at the time the processing for detection of an abnormality is performed, a pressures at the inside of the fuel tank and inside of the canister detected by the pressure sensor at every fixed time are stored in a storage device,
- a learned neural network learned in weights using the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the system causing leakage of fuel vapor as a truth label is stored, and
- at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak is detected from said input parameters by using the learned neural network.
2. The abnormality detection device of a fuel vapor escape prevention system according to claim 1, wherein the processing for detection of an abnormality includes processing for generating a valve opening instruction making the purge control valve open after generating a valve closing instruction of the purge control valve, a learned neural network learned in weights using the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where when perforation occurs in the system causing leakage of fuel vapor, a case where a valve opening abnormality occurs in which the purge control valve continues opened, and a case where a valve closing abnormality occurs in which the purge control valve continues closed as truth labels, respectively, is stored, and, at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak, a valve opening abnormality of the purge control valve, and a valve closing abnormality of the purge control valve are detected from the input parameters by using the learned neural network.
3. The abnormality detection device of a fuel vapor escape prevention system according to claim 1, wherein the input parameters are comprised of the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device, the atmospheric pressure when the processing for detection of an abnormality is performed, and a remaining amount of a fuel in the fuel tank when the processing for detection of an abnormality is performed.
4. The abnormality detection device of a fuel vapor escape prevention system according to claim 1, wherein the input parameters are comprised of the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device, the atmospheric pressure when the processing for detection of an abnormality is performed, a remaining amount of a fuel in the fuel tank when the processing for detection of an abnormality is performed, a temperature of the fuel in the fuel tank, and a parameter showing a capacity of the suction pump.
5. An abnormality detection device of a fuel vapor escape prevention system comprising:
- a canister formed with a fuel vapor chamber and atmospheric pressure chamber at the two sides of an activated carbon layer, the fuel vapor chamber being on the one hand communicated with an inside space above a fuel level of a fuel tank and on the other hand communicated through a purge control valve with an intake passage of an engine,
- a flow path switching valve able to selectively connect the atmospheric pressure chamber to the atmosphere and a suction pump, a passage from the flow path switching valve toward the atmospheric pressure chamber and a suction passage from the flow path switching valve toward the suction pump being connected by a reference pressure detection passage having a restricted opening, and
- a pressure sensor arranged in the suction passage from the flow path switching valve toward the suction pump,
- at the time of stopping operation of the vehicle, processing for detection of an abnormality is performed to generate a valve closing instruction making the purge control valve close, a pump operation instruction making the suction pump operate to make an inside of the fuel tank and inside of the canister a negative pressure while maintaining a switched position of the flow path switching valve at a switched position where the atmospheric pressure chamber is connected to the atmosphere when a predetermined time elapses after stopping operation of the vehicle, a switching instruction switching the switched position of the flow path switching valve to a switched position at which the atmospheric pressure chamber is connected to the suction pump after generation of the pump operation instruction, and a valve opening instruction making the purge control valve open after the generation of the switching instruction,
- at the time the processing for detection of an abnormality is performed, a pressures at the inside of the fuel tank and inside of the canister detected by the pressure sensor at every fixed time are stored in a storage device,
- a learned neural network learned in weights using the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the system causing leakage of fuel vapor as a truth label is stored, and,
- at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak is detected from said input parameters by using the learned neural network.
6. The abnormality detection device of a fuel vapor escape prevention system described in claim 5, wherein a learned neural network learned in weights using the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device and at least the atmospheric pressure when the processing for detection of an abnormality is performed as input parameters of the neural network and using a case where perforation occurs in the system causing leakage of fuel vapor, a case where a valve opening abnormality occurs in which the purge control valve continues opened, a case where a valve closing abnormality occurs in which the purge control valve continues closed, a case where an abnormality occurs in the pressure sensor, a case where a switching abnormality occurs in which the switched position of the flow path switching valve is maintained at a switched position connecting the atmospheric pressure chamber to the atmosphere, a case where a switching abnormality occurs in which the switched position of the flow path switching valve is maintained at a switched position connecting the atmospheric pressure chamber to the suction pump, a case where an abnormality occurs in which the suction pump continues operating, and a case where an abnormality occurs in which the suction pump continues stopped, respectively, as truth labels is stored, and
- at the time of stopping operation of the vehicle, a perforation abnormality causing fuel vapor to leak, the valve opening abnormality of the purge control valve, the valve closing abnormality of the purge control valve, an abnormality of the pressure sensor, the switching abnormality of the flow path switching valve, and an abnormality of the suction pump are detected from the input parameters by using the learned neural network.
7. The abnormality detection device of a fuel vapor escape prevention system according to claim 5, wherein the input parameters are comprised of the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device and the atmospheric pressure when the processing for detection of an abnormality is performed and a remaining amount of a fuel in the fuel tank when the processing for detection of an abnormality is performed.
8. The abnormality detection device of a fuel vapor escape prevention system according to claim 5, wherein the input parameters are comprised of the pressures at the inside of the fuel tank and inside of the canister at every fixed time stored in the storage device and the atmospheric pressure when the processing for detection of an abnormality is performed, a remaining amount of a fuel in the fuel tank when the processing for detection of an abnormality is performed, a temperature of the fuel in the fuel tank, and a parameter showing a capacity of the suction pump.
5657737 | August 19, 1997 | Ishida |
5781700 | July 14, 1998 | Puskorius |
5954783 | September 21, 1999 | Yamaguchi |
7181334 | February 20, 2007 | Kropinski |
7386388 | June 10, 2008 | Akazaki |
8818676 | August 26, 2014 | Peng |
20200123990 | April 23, 2020 | Yokoyama |
20200125042 | April 23, 2020 | Oyagi |
20200143607 | May 7, 2020 | Kitagawa |
2004-044396 | February 2004 | JP |
Type: Grant
Filed: Feb 18, 2020
Date of Patent: Nov 10, 2020
Assignee: TOYOTA JIDOSHA KABUSHIKI KAISHA (Toyota)
Inventors: Harufumi Muto (Miyoshi), Akihiro Katayama (Toyota), Yosuke Hashimoto (Nagakute), Kazuki Tsuruoka (Toyota)
Primary Examiner: Carl C Staubach
Application Number: 16/792,952
International Classification: F02M 25/08 (20060101); G05B 13/02 (20060101);